AGP speed 2X,4x,8X: What it really means!

[Guest]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

The number only refers to the transfer rate of the AGP bus, which is an
extension of the PCI bus. Yes, faster is better but it primarily aides
in reducing load time of the texture data to the video cards local
memory therefore frames per sec increase is negligible at best. 3D
Benchmark programs may show improved scores but that is due to the
reduced latency of the data getting to the card.
The reason the AGP bus was invented was increase performance with older
cards when Video RAM prices were high and they had 4MB~16MB. The idea
was to use some of the System RAM to store the data that the video card
couldn't hold then transfer that data to it directly with a minimum of
CPU usage. The aperture size setting was to define how much of the
system RAM could be used for this purpose. Modern cards with all their
RAM and processing power no longer benefit much if at all save the
transfer speed to reduce load times hence the switch to PCI-express.

 

[Guest]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

Chris Madsen wrote:

> The number only refers to the transfer rate of the AGP bus, which is an
> extension of the PCI bus. Yes, faster is better but it primarily aides
> in reducing load time of the texture data to the video cards local
> memory therefore frames per sec increase is negligible at best. 3D
> Benchmark programs may show improved scores but that is due to the
> reduced latency of the data getting to the card.
> The reason the AGP bus was invented was increase performance with older
> cards when Video RAM prices were high and they had 4MB~16MB. The idea
> was to use some of the System RAM to store the data that the video card
> couldn't hold then transfer that data to it directly with a minimum of
> CPU usage. The aperture size setting was to define how much of the
> system RAM could be used for this purpose. Modern cards with all their
> RAM and processing power no longer benefit much if at all save the
> transfer speed to reduce load times hence the switch to PCI-express.

PCI Express is considerably faster than AGP. The switch is mostly about
marketing.

--
--John
Reply to jclarke at ae tee tee global dot net
(was jclarke at eye bee em dot net)

 

[chip]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

"Chris Madsen" <take_no_shit@isp.com> wrote in message
news:cidcpc$akk$1@news.chatlink.com...
> The number only refers to the transfer rate of the AGP bus, which is an
> extension of the PCI bus.

You neglect to say that AGP is 64bits wide and runs at 133MHz (at 1X),
compared to 32bits and 33MHz for PCI. So even AGP 1X is 8 times faster than
PCI. People might be confused into thinking that AGP 8X is only 8 times
faster than PCI. This is not true: it is 64 times faster.

>Yes, faster is better but it primarily aides
> in reducing load time of the texture data to the video cards local
> memory therefore frames per sec increase is negligible at best.

[snip]

> Modern cards with all their
> RAM and processing power no longer benefit much if at all save the
> transfer speed to reduce load times hence the switch to PCI-express.

I don't agree here. The performance of PCI-express is not dramatically
better than AGP 8X. And since the performance increase you see when going
from AGP 4x to AGP 8x is very very small, expect a similarly pathetic
performance boost moving to PCI-express. Its principle advantages are for
Raid disk controllers and the like, because the old PCI standard has become
a real bottleneck; AGP has not.

With respect to graphics controllers, the only real advantages PCI-Express
offers are that (A) it is bi-directional. This is beneficial in some CAD
type modelling environments. and (B) the PCI-Express connector standard can
carry more current than the PCI connector. Bottom line is PCI-express is a
complete waste of time for gaming speed improvements. Anyone looking to
"upgrade" their graphics card specifically to take advantage of PCI-Express
is going to be very disappointed.

Chip

 

[Guest]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

>You neglect to say that AGP is 64bits wide and runs at 133MHz (at 1X),
>compared to 32bits and 33MHz for PCI. So even AGP 1X is 8 times

AGP runs at 66Mhz at 1X not 133Mhz. 133Mhz is 2X

 

[Guest]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

Chip wrote:
>
> The performance of PCI-express is not dramatically
> better than AGP 8X. And since the performance increase you see when going
> from AGP 4x to AGP 8x is very very small, expect a similarly pathetic
> performance boost moving to PCI-express. Its principle advantages are for
> Raid disk controllers and the like, because the old PCI standard has become
> a real bottleneck; AGP has not.
> {snip}
> With respect to graphics controllers, the only real advantages PCI-Express
> offers are that (A) it is bi-directional. This is beneficial in some CAD
> type modelling environments. and (B) the PCI-Express connector standard can
> carry more current than the PCI connector. Bottom line is PCI-express is a
> complete waste of time for gaming speed improvements. Anyone looking to
> "upgrade" their graphics card specifically to take advantage of PCI-Express
> is going to be very disappointed.

I'm not sure about the CAD application benefits but the old PCI bus is
still fast enough for RAID as its capacity is 133MB/s and the best RAID
speed will only burst at ~85MB/s in real world. Another big difference
is that PCI-express is a "serial" connection not "parallel" like AGP or
standard PCI and serial devices can invariably be clocked much higher
for better throughput. Power shouldn't be much of an issue as most
hungry cards have connectors on them to get what they need directly from
the main supply anyway.

Chris

 

[chip]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

"PRIVATE1964" <private1964@aol.com> wrote in message
news:20040917190132.19486.00000779@mb-m22.aol.com...
> >You neglect to say that AGP is 64bits wide and runs at 133MHz (at 1X),
> >compared to 32bits and 33MHz for PCI. So even AGP 1X is 8 times
>
> AGP runs at 66Mhz at 1X not 133Mhz. 133Mhz is 2X

Of course it does. What on earth was I thinking of. Doh!

Chip

 

[Guest]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

> With respect to graphics controllers, the only real advantages
> PCI-Express offers are that (A) it is bi-directional. This is
> beneficial in some CAD type modelling environments. and (B) the

The other advantage to PCI express is having multiple slots. It always
irked me that if I upgraded my AGP card, I couldn't run both at the same
time. Now if your current card is slow with new games, but has great VIVO
features, you won't need to look for another card that also does VIVO when
shopping for a new card.

 

[Guest]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

>Of course it does. What on earth was I thinking of. Doh!
>

It happens to the best of us, but I have to admit very rarely with myself do I
slip up. ; )

 

[Guest]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

>"serial" connection not "parallel"

I really don't understand that concept. How can data be faster moving down a
single path? Is it because the path can be run at a much higher clock speed?
What happens with parrallel does the data get corrupted easier?

 

[Guest]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

PRIVATE1964 wrote:
>>"serial" connection not "parallel"
>
>
> I really don't understand that concept. How can data be faster moving down a
> single path? Is it because the path can be run at a much higher clock speed?
> What happens with parrallel does the data get corrupted easier?

The speed difference between serial and parallel is; a byte needs to be
converted twice as often when its transmitted in parallel than when its
done serially. The main reason most things are done in parallel is
because its more cost effective and is easier to implement. Serial data
paths can't be too long because of resistance of the conductor which
causes data loss, thats why they use multiple shorter paths instead. The
main problem comes when there is too many of those shorter paths close
together and have "cross-talk" which causes data corruption. If any data
is lost or corrupt it needs to be re-sent and reduces throughput. This
happens more often with a parallel connection than serial, thats why
they needed the 80 conductor cables for IDE to support higher
throughput. Every other conductor is tied to ground to reduce cross-talk
between adjacent data lines. Hope this clears things up a bit.

Chris

 

[Guest]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

>
>The speed difference between serial and parallel is; a byte needs to be
>converted twice as often when its transmitted in parallel than when its
>done serially. The main reason most things are done in parallel is
>because its more cost effective and is easier to implement. Serial data
>paths can't be too long because of resistance of the conductor which
>causes data loss, thats why they use multiple shorter paths instead. The
>main problem comes when there is too many of those shorter paths close
>together and have "cross-talk" which causes data corruption. If any data
>is lost or corrupt it needs to be re-sent and reduces throughput. This
>happens more often with a parallel connection than serial, thats why
>they needed the 80 conductor cables for IDE to support higher
>throughput. Every other conductor is tied to ground to reduce cross-talk
>between adjacent data lines. Hope this clears things up a bit.
>
>Chris

It does thank you.

 

[Guest]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

"PRIVATE1964" <private1964@aol.com> wrote in message
news:20040918150559.01015.00000777@mb-m15.aol.com...
> >
> >The speed difference between serial and parallel is; a byte needs to be
> >converted twice as often when its transmitted in parallel than when its
> >done serially. The main reason most things are done in parallel is
> >because its more cost effective and is easier to implement. Serial data
> >paths can't be too long because of resistance of the conductor which
> >causes data loss, thats why they use multiple shorter paths instead. The
> >main problem comes when there is too many of those shorter paths close
> >together and have "cross-talk" which causes data corruption. If any data
> >is lost or corrupt it needs to be re-sent and reduces throughput. This
> >happens more often with a parallel connection than serial, thats why
> >they needed the 80 conductor cables for IDE to support higher
> >throughput. Every other conductor is tied to ground to reduce cross-talk
> >between adjacent data lines. Hope this clears things up a bit.
> >
> >Chris
>
> It does thank you.

I think it's parallel communication that can't be used for long distances
rather than serial. With multiple lines of data, the signals start to skew
with long distances (where the data bits arrive at different times instead
the same time like they're supposed to). And if you try to clock it faster,
the skew becomes a big problem. With serial communication, the data bits
will arrive in the same order that you send them no matter how fast it's
going. I guess that's one reason why SATA came about.

Vu.

 

[chip]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

"Chris Madsen" <take_no_shit@isp.com> wrote in message
news:cih6qv$bkm$1@news.chatlink.com...
> Chip wrote:
> >
> > The performance of PCI-express is not dramatically
> > better than AGP 8X. And since the performance increase you see when
going
> > from AGP 4x to AGP 8x is very very small, expect a similarly pathetic
> > performance boost moving to PCI-express. Its principle advantages are
for
> > Raid disk controllers and the like, because the old PCI standard has
become
> > a real bottleneck; AGP has not.
> > {snip}
> > With respect to graphics controllers, the only real advantages
PCI-Express
> > offers are that (A) it is bi-directional. This is beneficial in some
CAD
> > type modelling environments. and (B) the PCI-Express connector standard
can
> > carry more current than the PCI connector. Bottom line is PCI-express
is a
> > complete waste of time for gaming speed improvements. Anyone looking to
> > "upgrade" their graphics card specifically to take advantage of
PCI-Express
> > is going to be very disappointed.
>
> I'm not sure about the CAD application benefits but the old PCI bus is
> still fast enough for RAID as its capacity is 133MB/s and the best RAID
> speed will only burst at ~85MB/s in real world.

That's just not true I am afraid. I agree that often it is the case, but
certainly its not always the case. 2 x 74GB Raptors will transfer 140MB/s
in real world tests. (And BTW the maximum you seem to be able to get
through PCI is only around 117MB/s: although the spec allows 133, that's not
really available in practice.) Also, you completely ignore multiple disk
Raid systems. Since - broadly speaking - 4 x 100GB is not much different in
price from 1 x 400GB, Raid 5 becomes a real possibility for even home PC's.
Why not buy 4 smaller disks instead of 1 large one? Its much faster than a
single disk and offers resiliance too. A Raid 5 setup like this just can't
work properly on PCI.

[snip]

> Power shouldn't be much of an issue as most
> hungry cards have connectors on them to get what they need directly from
> the main supply anyway.

I agree its not much of an issue. But its a fudge, and its only been
brought about by the very limited power supply capability of the PCI slot.
The new PCI-Express standard aims to improve that. Certainly there will
always be some cards that will still need their own dedicated power supply,
but there will be many "mid-range" cards that won't need it on PCI-Express.
And this will help to keep the costs down and simplify wiring etc. I know
its not a big deal, but it is a marginal improvement.

Chip

 

[Guest]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

PRIVATE1964 wrote:

>>
>>The speed difference between serial and parallel is; a byte needs to be
>>converted twice as often when its transmitted in parallel than when its
>>done serially. The main reason most things are done in parallel is
>>because its more cost effective and is easier to implement. Serial data
>>paths can't be too long because of resistance of the conductor which
>>causes data loss, thats why they use multiple shorter paths instead. The
>>main problem comes when there is too many of those shorter paths close
>>together and have "cross-talk" which causes data corruption. If any data
>>is lost or corrupt it needs to be re-sent and reduces throughput. This
>>happens more often with a parallel connection than serial, thats why
>>they needed the 80 conductor cables for IDE to support higher
>>throughput. Every other conductor is tied to ground to reduce cross-talk
>>between adjacent data lines. Hope this clears things up a bit.
>>
>>Chris
>
> It does thank you.

Except that it makes no sense.

--
--John
Reply to jclarke at ae tee tee global dot net
(was jclarke at eye bee em dot net)

 

[Guest]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

Chris Madsen wrote:

> PRIVATE1964 wrote:
>>>"serial" connection not "parallel"
>>
>>
>> I really don't understand that concept. How can data be faster moving
>> down a single path? Is it because the path can be run at a much higher
>> clock speed?

Got it in one.

>> What happens with parrallel does the data get corrupted
>> easier?

It can if the circuit is badly designed.

The real purpose of PCI Express though is to make money for Intel. There
are alternatives that give the same performance and allow the use of
existing PCI boards in the same slots that take the higher performance
boards. But that would mean that people didn't have to throw away there
entire investment in older hardware to use the new technology.

> The speed difference between serial and parallel is; a byte needs to be
> converted twice as often when its transmitted in parallel than when its
> done serially.

"Converted"? From what to what? And why would serial take fewer
"conversions" than parallel?

> The main reason most things are done in parallel is
> because its more cost effective and is easier to implement.

"Most things"? Such as?

> Serial data
> paths can't be too long because of resistance of the conductor which
> causes data loss, thats why they use multiple shorter paths instead.

So how does using parallel data paths gain you anything over serial if the
parallel paths are shorter? And it's possible to transfer about the same
amount of over 100 meters of CAT5 cable using gigabit Ethernet as over the
few inches of parallel circuit trace that makes up the PCI bus, so what's
wrong with your picture?

> The
> main problem comes when there is too many of those shorter paths close
> together and have "cross-talk" which causes data corruption. If any data
> is lost or corrupt it needs to be re-sent and reduces throughput.

I see, so what mechanism on the PCI bus resends corrupted data?

> This
> happens more often with a parallel connection than serial,

All else being equal it does. It's debatable whether running a serial
connection 16 times faster gives any real improvement in reliability
though.

> thats why
> they needed the 80 conductor cables for IDE to support higher
> throughput. Every other conductor is tied to ground to reduce cross-talk
> between adjacent data lines.

The first part that you've gotten unambiguously right.

> Hope this clears things up a bit.
>
> Chris

--
--John
Reply to jclarke at ae tee tee global dot net
(was jclarke at eye bee em dot net)

 

[Guest]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

>Except that it makes no sense.

It makes sense.

 

[Guest]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

PRIVATE1964 wrote:

>>Except that it makes no sense.
>
> It makes sense.

Only if you don't try to apply logic to it.

--
--John
Reply to jclarke at ae tee tee global dot net
(was jclarke at eye bee em dot net)

 

[chip]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

"PRIVATE1964" <private1964@aol.com> wrote in message
news:20040921220140.29600.00001117@mb-m29.aol.com...
> >Except that it makes no sense.
>
> It makes sense.

Sorry Private, I am with J Clark on this one. I think you have some useful
concepts, but some of it is not quite right. For example this "byte needs
to be converted twice..." stuff. That's complete nonsense. The data is
parallel in the first pace. The CPU reads and writes words, not bits, or
even bytes. Its very easy for a parallel controller to read data in
parallel and output it in parallel. Ironically, to do it serially takes
more work.

Chip.

 

[Guest]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

>> It makes sense.

I meant it makes sense in a general way.

Like the extra 40 conductors in an EIDE cable which help keep crosstalk down.
That's accurate. Length of cable also. EIDE is maxed at 18" I believe but the
serial cable could be longer and run faster because of less chance of cross
talk. It's a lot easier to shield a single cable then to shield a large number
of cables like with EIDE.

That's all I know and I'm sticking by it. : )

 

[Guest]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

Like the extra 40 conductors in an EIDE cable which help keep crosstalk down.
That's accurate. Length of cable also. EIDE is maxed at 18" I believe but the
serial cable could be longer and run faster because of less chance of cross
talk. It's a lot easier to shield a single cable then to shield a large number
of cables like with EIDE.

That's all I know and I'm sticking by it. : )

--------------------------------------------------------------------------
-------------------
I'm running a hard drive on a 36" round IDE cable with no problems. I needed it
because it's a full tower case so the cable will reach. DOUG

 

[Guest]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

Courseyauto wrote:

> Like the extra 40 conductors in an EIDE cable which help keep crosstalk
> down. That's accurate. Length of cable also. EIDE is maxed at 18" I
> believe but the serial cable could be longer and run faster because of
> less chance of cross talk. It's a lot easier to shield a single cable then
> to shield a large number of cables like with EIDE.

If serial cable is necessary to get 1 meter out of 150 MB/sec ATA, then how
do they manage to run U320 SCSI over 25 meters of parallel cable?

Sorry, but that argument doesn't wash.

Further, SATA cables are not shielded.

PATA goes 18" because that's what the spec said and the chip designers
worked to the spec. SATA goes one meter because that's what the spec said
and the chip designers worked to the spec. U320 SCSI goes 25 meters
because that's what the spec said and the chip designers designed to the
spec. If a new PATA spec came out that required 200 MB/sec over 30 meters
on 40-wire parallel cable then the chip designers would probably have found
a way to make it work.

>
> That's all I know and I'm sticking by it. : )
>
> --------------------------------------------------------------------------
> -------------------
> I'm running a hard drive on a 36" round IDE cable with no problems. I
> needed it
> because it's a full tower case so the cable will reach. DOUG

--
--John
Reply to jclarke at ae tee tee global dot net
(was jclarke at eye bee em dot net)

 

[Guest]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

>I'm running a hard drive on a 36" round IDE cable with no problems. I needed
>it
>because it's a full tower case so the cable will reach. DOUG

The 18" is just a spec, it doesn't mean longer won't work but there is a risk
of having data corruption.

 

[Guest]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

>f a new PATA spec came out that required 200 MB/sec over 30 meters
>on 40-wire parallel cable then the chip designers would probably have found
>a way to make it work.

I agree, but then you would still have the wide cable.

 

[Guest]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

PRIVATE1964 wrote:

>>f a new PATA spec came out that required 200 MB/sec over 30 meters
>>on 40-wire parallel cable then the chip designers would probably have
>>found a way to make it work.
>
> I agree, but then you would still have the wide cable.

And that's the only real benefit of SATA that I can see that is inherent in
its being serial--that narrow cable is a lot easier to route.

--
--John
Reply to jclarke at ae tee tee global dot net
(was jclarke at eye bee em dot net)

 

[Guest]

Archived from groups: alt.comp.periphs.videocards.nvidia (More info?)

>
>And that's the only real benefit of SATA that I can see that is inherent in
>its being serial--that narrow cable is a lot easier to route.

Here's a question for you that will keep me from having to search on related to
SATA. I plan on getting a new serial hard drive soon.

I'm using a NF7-S with serial connections.
What is the maximum throughput for a hard drive connected to the serial
connection.

Is that connection spec'd for 150Mb/sec? I've read that the serial drives that
are out now are not "true native" serial drives so there is no way they could
ever hit 150Mb/sec.

What can you tell me about this please.

Thanks